Research On Shot Boundary Detection Method

Dividing a video sequence in to shots is the first step towards video-content analysis and content-based video detection and retrieval. This thesis addresses the key technical problems arising in shot boundary detection and video caption detection, proposes the new methodologies and algorithms for them. This paper analyzes the difference of different shot change, then based of the difference feature we class the shots into gradual shots and cut shot, an improved algorithm for shot segmentation is proposed to detect the shot., the main work as follows:1. According to the change of features, we design peak-selector; using peak-selector to decide do abrupt shot cuts detection or gradual shot cuts detection in order to achieve the best possible detection performances.2. To detect the cut shot, Frame sequence's in-continuity or false segmentation can be caused by unsupervised clustering, and the existing techniques are based on sequential search, while is too expensive for practical use. To solve the problems above, an improved algorithm for shot segmentation based on clustering is proposed, a referenced variable is used in clustering to decide whether the shots continue clustering, and most video frames do not need to be compared .It can improve their performance substantially.3. To the gradual shot such as fade-in, fade-out and dissolve, we should divide the image into 16×16 macroblocks, and then compare the gray value of the edge in the macroblocks between different frames at the same location. According to the change of the features in gradual shot, we design GSC detector. The simulation results show that the proposed algorithm can segment whole video object accurately.All the algorithms have been realized in the platform VC6.0. Finally, the evaluations and test results are presented.